[object Object]

[object Object][object Object]undefined[object Object]

• Description: Traverses different sampling points of feature maps of different sizes by using parameters such as the sample location, attention weights, mapped value feature, start index location of a multi-scale feature, and spatial size of a multi-scale feature map (which facilitates changing a sampling location from a normalized value to an absolute location).

• Formula:

  Map the normalized coordinates $(u,v)\in[0,1]$ of the sampling point to the pixel coordinate system of the feature map at layer $\ell$:

  $$x = u \cdot W_\ell - 0.5, \qquad y = v \cdot H_\ell - 0.5$$

  Determine the four integer grid points between which the sampling point falls:

  $$ x_0 = \lfloor x \rfloor,\quad x_1 = x_0 + 1,\qquad y_0 = \lfloor y \rfloor,\quad y_1 = y_0 + 1 $$ Compute the offset of the sampling point relative to the upper-left grid point, which is used for interpolation weighting: $$

  \alpha_x = x - x_0, \qquad \alpha_y = y - y_0 $$

  Compute the bilinear interpolation weight. The sum of the four adjacent points is 1.

  $$\begin{aligned} w_{00} &= (1-\alpha_y)(1-\alpha_x), \\ w_{10} &= (1-\alpha_y)\alpha_x, \\ w_{01} &= \alpha_y(1-\alpha_x), \\ w_{11} &= \alpha_y\alpha_x \end{aligned}$$

  Compute the feature vectors (length: $D$) corresponding to the sampling points.

  $$ \operatorname{bilinear}(V;\,b,h,\ell,x,y) = w_{00} \, V_{b,\ell,y_0,x_0,h,:} + w_{10} \, V_{b,\ell,y_0,x_1,h,:} + w_{01} \, V_{b,\ell,y_1,x_0,h,:} + w_{11} \, V_{b,\ell,y_1,x_1,h,:} $$ Compute the weighted sum of the bilinear sampling results for all layers and all sampling points to obtain the final output: $$

  O_{b,q,h,:} = \sum_{\ell=0}^{L-1} \sum_{p=0}^{N_p-1} A_{b,q,h,\ell,p} \cdot \operatorname{bilinear}!\left(V;,b,h,\ell, x_{b,q,h,\ell,p}, y_{b,q,h,\ell,p}\right) $$

[object Object]

Each operator has calls. First, aclnnMultiScaleDeformableAttnFunctionGetWorkspaceSize is called to obtain the workspace size required for computation and the executor that contains the operator computation process. Then, aclnnMultiScaleDeformableAttnFunction is called to perform computation.

[object Object]

• Parameters:

  [object Object]
  • Atlas inference series products: BFLOAT16 is not supported.

• Returns:

  aclnnStatus: status code. For details, see .

  The first-phase API implements input parameter verification. The following errors may be thrown:

  [object Object]

[object Object]

• Parameters:

  [object Object]

• Returns:

  aclnnStatus: status code. For details, see .

[object Object]

• Deterministic compute:

  • aclnnMultiScaleDeformableAttnFunction defaults to a deterministic implementation.

• [object Object]Atlas inference series products[object Object]:

  • channels%32 = 0, and channels ≤ 256
  • 32 ≤ num_queries < 500000
  • num_levels ≤ 16
  • num_heads = [2, 4, 8]
  • num_points = [4, 8]

• [object Object]Atlas A2 training products/Atlas A2 inference products[object Object] and [object Object]Atlas A3 training series products/Atlas A3 inference series products[object Object]:

  • channels%8 = 0, and channels ≤ 256
  • 32 ≤ num_queries < 500000
  • num_levels ≤ 16
  • num_heads ≤ 16
  • num_points ≤ 16

[object Object]

The following example is for reference only. For details, see .